Re: st: y-standardisation in logistic multilevel models with gllamm

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Wed, 16/3/11, Tina Schmid wrote:
> When comparing coefficient X1 in binary logistic regression
> across models (same sample, but different covariates), y
> standardization of the coefficient is required (see for
> example Mood 2010, Logistic Regression: Why we cannot do
> what we think we can do, and what we can do about it, ESR,
> 26 (1), 67-82). So far, so good.

Unfortunately, that is not quite true (and it is also not
quite the conclusion from the article that you cite). 
Standardization will only help when you are interested in the
standardized latent dependent variable, which is almost never 
the case. More often you are interested in either the 
probabilty or odds, and these change in fundamental ways when
you add or remove variables. Probabilties or odds contain both
information on the central tendency and on uncertainty. The 
uncertainty comes from variables not in your model. So the 
definition of our dependent variable depends on the variables
we choose to include in our model, and when we change those
variables then we are also changing the definition of our 
dependent varianble, making it impossible to directly compare 
coefficients. Standardization won't help, as that is working
on a different scale. One possibility is to choose one of the 
models as the definition of probability of interest and 
transform the remaining model to that scale. This is what I 
did in:

Maarten L. Buis (2010) "Direct and indirect effects in a logit
model" The Stata Journal,10(1): 11-29.

Botom line is that there is no easy solution to this problem.

Hope this is not too depressing,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/